Message boards : Graphics cards (GPUs) : SLI - GPUGRID GTX980
| Author | Message |
|---|---|
 @tonymmorleySend message Joined: 10 Mar 14 Posts: 24 Credit: 1,215,128,812 RAC: 0 Level  Scientific publications     | |
|
Easy question, to SLI or not? | |
ID: 41589 | Rating: 0 | rate:
 /   Reply
Quote
| |
|
skgiven Volunteer moderator Volunteer tester  Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level  Scientific publications   | |
|
Not. | |
| ID: 41590 | Rating: 0 | rate:
/ Reply
Quote
| |
|
@tonymmorley Send message Joined: 10 Mar 14 Posts: 24 Credit: 1,215,128,812 RAC: 0 Level Scientific publications | |
|
I presume it's because if one card slows down it's going to pull down the second card? | |
| ID: 41591 | Rating: 0 | rate:
/ Reply
Quote
| |
Retvari Zoltan Send message Joined: 20 Jan 09 Posts: 2343 Credit: 16,206,655,749 RAC: 261,147 Level  Scientific publications | |
|
No, SLI can cause various computing errors, and it's practical to reduce any risks to avoid loosing hours of crunching. The next CUDA (7.5) release probably will support SLI crunching, but the app should be updated to this CUDA release to utilize this feature. But it's unnecessary from the project's point of view, so don't expect it any time soon. | |
| ID: 41592 | Rating: 0 | rate:
/ Reply
Quote
| |
|
SLI also mirrors the graphics card memory between cards, and likewise with multiple GPU cards such as Titan Zs. So for example, if you run one task per GPU and have two cards in SLI, both cards will store both tasks in their memory even though only one is being worked on by a particular GPU. This effectively doubles the memory usages and reduces the capacity potential needlessly. | |
| ID: 41741 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Retvari Zoltan Send message Joined: 20 Jan 09 Posts: 2343 Credit: 16,206,655,749 RAC: 261,147 Level Scientific publications | |
|
Even if SLI crunching will be supported from both parties (NVidia & GPUGrid app), I expect that the crunching speed won't be doubled (tripled, quadrupled etc), just like the frame rendering speed of different 3D engines (gaming, CAD/CAM, VR, etc) doesn't scale in direct ratio of the number of SLIed GPUs. | |
| ID: 41742 | Rating: 0 | rate:
/ Reply
Quote
| |
Even if SLI crunching will be supported from both parties (NVidia & GPUGrid app), I expect that the crunching speed won't be doubled (tripled, quadrupled etc), just like the frame rendering speed of different 3D engines (gaming, CAD/CAM, VR, etc) doesn't scale in direct ratio of the number of SLIed GPUs. I agree. It is invariably more efficient to just run additional tasks on additional GPUs (which can already be done) than to try managing the dissemination and reintegration of work by running a task on multiple GPUs (while also "wasting" card memory). In my case, my cards are in SLI, but GPUGrid doesn't make use of SLI, nor should it. Fortunately, my cards have a relatively large amount of memory, so wasting it on having task memory usage mirrored between the cards isn't really an issue for me. The only thing I can think of where SLI might potentially be significantly useful to projects like GPUGrid is if it could somehow be used for some kind of on-the-fly error correction as apposed to increasing the amount of work per time done. That's just an off-the-wall hypothetical though. ____________ My BOINC Cruncher, Minecraft Multiserver, Mobile Device Mainframe, and Home Entertainment System/Workstation: http://www.overclock.net/lists/display/view/id/4678036#  | |
| ID: 42150 | Rating: 0 | rate:
/ Reply
Quote
| |
Message boards : Graphics cards (GPUs) : SLI - GPUGRID GTX980
